差別化雲端運算服務之定價模型

Abstract

本論文根據契約理論與賽局理論提出兩種定價模式，這兩種定價模式分別適用於不同情形。差異化定價機制適用於雲端運算剛風行而未普及時，此時顧客的私有運算資源的多寡成為雲端運算廠商關注的焦點。當顧客的私有運算資源愈多，他愈不願意付很多錢購買雲端運算服務；當顧客的私有運算資源很少， 他願意付出較多金錢購買雲端運算服務以得到額外的運算資源。為此，我們設計差異化定價機制來自動區分出顧客私有運算資源的多寡。此機制是廠商先提供菜單給顧客，菜單上標示著不同的單價及其相對應之數量，顧客在看到菜單後，會選出對自己有利的數量及單價來購買。差異化定價機制不僅可以區分出不同類型的顧客， 亦可以最大化廠商的利潤。一般定價機制適用於雲端運算完全普及時，此時顧客的私有運算資源對比於其所購買的雲端運算資源應是很少，故使用差異化定價機制後廠商可能得出很少差異的差異化定價，較無實益。因此，廠商應著重於一般定價機制。在此模型中，雲端運算資源被分成兩類，一類是需求型資源（on-demand instances）， 一類是點狀資源（spot instances）。需求型資源是採取固定單價來販賣，點狀資源是顧客投標決定價格。我們考慮廠商先賣需求型資源再把剩下的資源以點狀資源的型式出售。此一般定價機制有兩個重要參數：需求型資源的價格與欲販售之點狀資源的數量。本論文設計這兩個參數，以符合廠商的最大利益。

We propose pricing mechanisms for cloud computing systems based on contract theory and game theory. Differentiated pricing mechanism is for the intermediate period in which private data center is still popular. A service provider needs to deal with customers with and without private data center. Those customers without private data center tend to buy more cloud computing resources while customers with private data center use cloud computing only when it is cheap, due to the concept of decreasing marginal value in Economics. As a result, a service provider needs to know if customers have private data center for the purpose of price discrimination. In differentiated pricing mechanism, a service provider provides a menu for customers to order virtual machines, which is similar to the mechanism in Amazon Elastic Compute Cloud. We aim to design a menu that can maximize service provider’s profit. To achieve this goal, we use principal-agent model to induce customers' hidden information, which is endowment computing power (i.e. size of private computing cloud). Our result shows that our differentiated pricing mechanism is able to maximize service provider’s profit and keep customer’s utility at their acceptable level. Differentiated pricing mechanism also shows that customers without her own private computing cloud will be exploited to get a lower utility gain than those with her own private computing cloud. Numerical results are also provided. General pricing mechanism is a general mechanism which can be applied to a system containing on-demand instances and spot instances like Amazon Elastic Compute Cloud. In the system, the service provider sells on-demand instances first. Then the service provider sells the remaining resources for spot instances. Our mechanism determines at which price on-demand instances should be sold and how many remaining resources should be sold for spot instances in order to maximize service provider’s revenue. Empirical data from Amazon Elastic Compute Cloud is also provided.